hospitalverse-mcp

hospitalverse-mcp

Enables management of hospital operations through MCP tools for bed allocation, staff assignment, equipment tracking, and event simulation, with conflict resolution and live audit trail.

Category
访问服务器

README

hospitalverse-mcp

Rule-based Layer 2 agent logic + Layer 1 tool surface for HospitalVerse AI, matching the architecture in HospitalVerse_Architecture.md. This is build-order step 2 ("deterministic agents producing the same JSON shape a real agent would") — swap in LangGraph + an LLM per module (step 3) without touching the tool contracts, then MCP-ify (step 4).

What's here vs. what's new

Everything in src/ implements the tool surface from architecture §4. Two things go beyond the original doc, from the review pass:

  1. Visible conflict resolution. coordinator.ts doesn't just chain agent outputs — when the Bed Agent's top pick lands in an understaffed ward, or a doctor is over capacity, the Coordinator picks between two real options and records both: what it rejected, what it chose, and why. That's what ActionStep.conflict carries through to the Explainer prompt and the dashboard.
  2. A queryable audit trail. audit.resources.ts exposes audit://trail/{agent} and audit://trail/by-event/{eventId} as MCP Resources, not just a database table. Every tool call already writes to the log via withAudit() in state/dts.ts — these resources are what let a judge (or a real compliance reviewer) query it live.

Run it standalone (no MCP, no Studio)

npm install
npm run dev

src/index.ts runs a CLI demo: injects an ambulance surge, runs the Coordinator, and prints the resulting action chain — including the conflict-resolution step — plus the audit log size. Good for verifying agent logic works before wiring up Studio.

Wire it into NitroStack Studio

The folder layout here matches what nitrostack-cli init scaffolds (src/index.ts, app.module.ts, modules/*.tools.ts), but doesn't depend on @nitrostack/core yet — src/lib/define-tool.ts is a thin stand-in. To go from "runs standalone" to "connects in Studio":

  1. nitrostack-cli init hospitalverse-mcp --template typescript-starter
  2. Copy this src/ tree into the scaffolded project, replacing the starter's calculator example.
  3. In src/lib/define-tool.ts, swap the plain defineTool/ defineResource functions for whatever @nitrostack/core exports — every module here only imports from that one file.
  4. npm run dev, then Studio → Add Server → Nitro Project → pick the folder (handbook §4). The Tools/Resources pages will list everything in app.module.ts automatically.
  5. For the "no user typed a question" wow-moment: deploy, then connect to ChatGPT (handbook, "How to connect to ChatGPT") and call sim.inject_event from a ChatGPT chat instead of a dashboard button — proves the MCP surface is a real interop layer, not just internal plumbing for your own UI.

Files

src/
├── index.ts                    # standalone CLI demo / bootstrap
├── app.module.ts                # registers every tool + resource
├── lib/define-tool.ts           # @nitrostack/core stand-in — swap here
├── state/
│   ├── dts.ts                   # Digital Twin State + event bus + audit hook
│   └── state.tools.ts           # state.get_snapshot, get_historical_trend
├── predict/predict.tools.ts     # predict.occupancy (Forecaster Agent)
├── beds/beds.tools.ts           # beds.*, patients.list_dischargeable
├── staff/staff.tools.ts         # staff.* (Staffing Agent)
├── equipment/equipment.tools.ts # equipment.* (Equipment Agent)
├── sim/sim.tools.ts             # sim.inject_event — demo trigger
├── audit/audit.resources.ts     # audit://trail/* — the new compliance surface
├── coordinator/
│   ├── coordinator.ts           # conflict-resolution logic, two cascades
│   └── explainer.prompts.ts     # Explainer Agent prompt + fallback
└── health/health.ts             # health check for Studio's Health page

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选